《X射线探测焊缝及机械损伤方法概述----中英文翻译.docx》由会员分享，可在线阅读，更多相关《X射线探测焊缝及机械损伤方法概述----中英文翻译.docx（12页珍藏版）》请在课桌文档上搜索。
1、Originaltext：X-RAYDETECTSWELDSANDMECHANICALSTRUCTUREDAMAGEMATHODS,SUMMARIZEThemovingsmallobjectdetectioninimageisalwaysadifficultprobleminfieldofimageprocessing,whichappliesinmanyfields,suchasindustrialdetectionandmedicaldetection.Thedefects,suchasblowholesandincompletepenetration,occasionallyappear
2、intheweldingprocess.Thesedefectscanaffectthequalityandthesecurityofproducts.Therefore,defectsdetectioninweldingseamisextremelyimportant.Now,theon-linedetectionofdefectsintheweldisstilldonebyhumaninterpreter.However,thisprocessissubjective,inconsistent,laborintensiveandfatigueofinterpreter.Itisdesira
3、bletofindaneffectiveautomaticdefectsdetectionmethodtoassisthumaninterpreterinevaluatingthequalityofweldandtomaketheon-linedetectionobjective,standardandintelligent.Ourresearchisbasedonthis.Wehavestudiedtheautomaticdefectsdetectionintheweldseamandmainlydonethefollowingresearch:(1)Thereismuchredundant
4、backgroundinformationforthedefectsdetectionintheimage.Thereforeweuseanautomaticallyabstractingmethodofweldareabasedontheauto-adaptedthresholdsegmentation.Thismethodcanreducethecomputationandincreasetheprecision.(2)TheSUSANalgorithmhasgoodanti-noiseability,whichcanrecognizetheimageedgeverywell.Soweha
5、vestudiedadefectsdetectionmethodbasedonSUSANalgorithm,whichassociatedwiththemorphologyoperation.Theresultsindicatethatthismethodiseffective.(3)Waveletanalysismethodhasaverygoodlocalizationcharacteristic,whichcanfocusonthearbitrarydetailoftheanalyzedobject.Therefore,Westudiedamethodusingwaveletdecomp
6、ositiontogettheshapeandpositioninformationofthedefects.Thenweusethewienerfilterandmorphologymethodtocompletethedetection.Theautomaticflawdetectionofweldedtubesisoneofthemostimportantstepstoensurethequalityofthetubes.Nondestructiveinspectiononweldingseamoftubeisrequiredinthetubeproduction,andrealtime
7、X-Rayradiographyinspectionisaneffectivemeans.Alongwithcontinuousimprovementoftheproductiveratio,thedemandfortheautomaticinspectiontotheweldingseambecomesmoreandmorepressing,soimplementationoftheautomaticinspectionpossessesimportantsignificanceonboththeoryandreality.Wavelettransformisapowerfultoolint
8、hesignalandimageprocessing,anditsfundamentaltheoryhasbeenformed.Fromtheviewofengineeringapplications,however,thewavelettransformisstillintheelementarystage,thefurtherResearchesarerequiredforthepracticaluses.Inthisthesis,Weconcentratemainlyonusingwaveletanalysisforweldingseamimageprocessingandrecogni
9、tion,andsomerelatedtechniquesaredeveloped.Forconstructingweldingseampositioninganddetectioncontrolsystem,themultiplecomputersconfigurationforweldseamimagerecognitionisproposed.ThesystemadoptsthearchitectureinwhichmultipleCPUsprocessparallelsunderthecontrolofthemasterIPCcomputer.Thesystemcanperformst
10、oringtheweldseamimages,positioning,flawsrecognizingandqualityprejudging.TheWatch-Doginterfacecardissuccessfullydeveloped;itcanimprovethesystemreliabilitybyredundanciestechniqueofsavingbreakpointdataandrestoringthem.ThehardwaresupportingthesystemmakesuseofthehighspeeddigitalsignalprocessorTMS320C30fr
11、omTaxaxInstrumentsCompany.Theframegrabbercancapture25framesofweldingseamimagepersecondcontinuouslyandmakeitpossibletofulfilltherealtimeweldingseamimageprocessingAndrecognition.TheonekindofimprovedFWT(FastWaveletTransform)algorithmforafinitesequenceisproposedafterstudyingtheoryofmustiersolutionanalys
12、isandanalyzingtechnicalcharacteristicsofDSP.TheimplementationoftheperiodicextensionoftheFWTonDSPisdescribedindetailandthecorrespondingFWTassemblycodeisdescribedfortheDSPTMS320C3Xseries.Thisdissertationsuggestsschemeofimagedemonizingbasedontwo-dimensionaldiscretewavelettransform.Thedemonizingalgorith
13、misdescribedwithsomeoperators.Bythresholdthewavelettransformcoefficients,ofnoisyimages,theoriginalimagecanbereconstructedcorrectly.Differentthresholdselectionsandthresholdmethodsarediscussed.Thenewrobustlocalthresholdschemeisproposed.Quantifyingtheperformanceofimagedemonizingschemesbyusingthemeansqu
14、areerror,theperformanceoftherobustlocalthresholdschemeisdemonstratedandiscomparedwiththeuniversalthresholdscheme.Theexperimentshowsthatimagedemonizingusingtherobustlocalthresholdperformsbetterthanthatusingtheuniversalthreshold.Inordertoimprovetheaccuracyandtherealtimeperformanceofedgedetection,ameth
15、odneedtobefoundtomatchthedetectionoflowcontrastblurredweldingseamimage.Thisdissertationanalyzedthemainsourcesofnoiseaswellasthedifferentcharacteristicsofnoiseandsignalunderwavelettransform,andproposedaMoultriesolutionedgedetectionmethodbasedonwavelettransform.Theexperimentalresultsshowtheeffectofthi
16、salgorithmisadvantageousoverthatoftraditionaledgedetectionalgorithm.Thegeometricalrelationofellipticimagingisstudiedforweldingseamimageofthebuttweldsinstraighttubes.Theregionmodelofweldingseamimageisproposed,Itfurnishesaevidencetheorytofurtherprocesstoweldingseamimage.Combiningwiththeregionmodel,amo
17、del-basedadaptivetargetsegmentationalgorithmisproposed.OnebasisofthealgorithmisOtsu,sdiscriminatescriterion.Theadaptivetargetsegmentationofweldingseamimageisrealized.Theeffectoftargetimagesegmentationisquitewell.Thedifficultproblemoftargetflawautomaticrecognitioninweldingseamimageisanalyzed.Usingfor
18、referencetheconsciousnessorganizingprocessofthehumanvisionsystem,aknowledge-basedtargetrecognitionalgorithmwithmufti-featurefusion,mufti-windowarchitectureandmustiersolutionispresented.Withthehelpofcertainpriorknowledge,criteriaandmeansofartificialintelligence,targetflawsareextractedandrecognizedqui
19、tewell.ItisaProspectingintelligentrecognitionalgorithm.Thefastfeatureextractionalgorithmfortargetgeometricalfeatureisproposed.Thisalgorithmisdifferentfromusualfeatureextractionmethodswhichfirstneedtochangeagrayimageintobinaryimage.Thealgorithmsproposedgetthefeatureofaimageinthegrayimagedirectly.Usin
20、gthisalgorithmcanfastextractfeaturesoftargetflawsinweldingseamimage.Allkindsofmechanicaldevicesandstructuretendtobecomelarge-scaleandhighefficientwiththeindustrydevelopingandprogressofscienceandtechnology.Themechanicaldevicesandstructurebecomeverycomplextomeettheneedofindustry.Thestructureordevicesa
21、redamagediscouldntavoidedduringworkingundercomplexloadandworkingforalongtime.Thelosscausedbycrash,fatigue,erodingandwearisabout6%一8%ofGDPofUSAandJapan.Inourcountry,accidentnumberofstructuraldamageis10timesasmanyasthatin,intenseindustrializationcountryineightiesoflastcentury.In1986,thelossis12hundred
22、million$causedbythespaceshuttlenamedschallengerofU.S.A,crashed.In1985,theaccidentcauseofjointofelectromotorsetofDatongpowerplanecrashed,In1988,theaccidentcauseofmainbeamofelectromotorsetofQianlongpowerplanecrashed,thoseaccidentcauseoflossnear1hundredmillionRMB.Inourcountry,6seriousaccidentswarebeenc
23、ausedbyrotorofover50Mwelectromotorsetdamagedbadlyduring1984to199).Therefore,thestudythetheoryandtechniqueaboutlargescaleandcomplexmechanicaldevicesandstructureonlineinspectandearlyfaultdiagnosisisurgenttask.Especially,howtodetectthefaultofstructureasearlyaspossibleisengineersmostwanttodo.Butitisvery
24、difficultthatfaintsignalproducedbyearlyfaultisrecognized.Theresearchreportsofourcountryandoverseasshowthatatpresentthestudyofstructuredamageinspectbyvibrationcharacteristicscarryoutmostonthesimpleandsymmetrystructurejustasbeamandframeetc.andresultisgivenbasedonfiniteelementnumericcalculation.Butthes
25、tructureandjoinofpracticaldeviceisquitecomplex,itisimpossiblemodelingthepracticaldevicereliablybyFE.Sothatrealizingcomplexindustrialdeviceinspectonlineandearlyfaultdiagnosisbyusingvibratetestingtechniqueisaproblemthatwantstobesolvedurgently.Faultdiagnosistechniqueisintercrosssubject.Especially,theba
26、seintheoryoffaultdiagnosisofcomplexsystemisprovidedbasedonmodemcontroltheory,signalprocessing,patternrecognition,optimummethod,decision-makingandmanualintelligentaredevelopedrapidly.Structuredamagedetectionisaresearchprojectthathaswidebackgroundofindustrialapplication.Butrealizinglarge-scaleandcompl
27、exstructuredamageinspectonlineisdependontechniquessuchasdevelopmentofaccuratetestingtechniqueandsignalprocessingmethod,basedongettingtothebestadvantagemixmodelofstructuredamagedetection,thesensorsescapeplacedonstructurereasonablyandoptimallyThelarge-scalevibratingdeviceasaresearchedobject,themethods
28、tructuredamagedetectionisstudied.IngeneralNDTtechniquesuchultrasonictest,raytest,magnetismappliedinmodelingofflinemostly.Theprojecttestandschemepervadetestetc.arevibratepropertiesandstructuredamagecharacteristicsfromplatitudinous:offlinetestsandanalysistostructureasimpotentinformationofonlineautomat
29、icfaultdiagnosisdatabase.Thenthemethodoffew-testing,pointsmodelingtogettingstructuredamageinformationhasbeerresearched.Placingsensorsreasonablyandrealizinglarge-scaleanComplexstructuredamageinspectonlinearetargetsofthisproject.Thelarge-scalevibratingscreenhasbeenappliedwidelyincoaxindustryandotherin
30、dustrialareasasakindofimportantdevice.A:vibrationmechanicaldevice,itworksveryhardlyandworksinverbwretchedenvironmentsothatthebeamofscreenisdamagedeasilyTherefore,itisveryimportanthowtodetectthefaultofbeamasearlyapossibletomaketherepairschedulereasonablyandeconomicallyandtoavoidthebodyhurtanddeviceda
31、mage.Inthisthesis,howcanlocateadamagedbeamofscreenisstudiedserially.Theregulationofbeamvibrationcharacteristicschangedependondamagedegreeofbeamisfound.Alsotheregulationofwholescreenvibrationcharacteristicschangedependondamagedegreeofbeamisfoundtoo.Basedondeepresearchaboutbeamvibrationcharacteristics
32、changeregulationandwholescreenvibrationcharacteristicschangeregulationinseries,wecangettheoptimumplacetoplacedsensorsforlocationwhichbeamisdamaged.Thetargetofthethesisiscombinetheon-linedynamicallyinspectscreenforstructuredamagewithaccuratelylocatefaultbyacousticemissiontechnique.Themaincontentofthi
33、sthesisconsistof(1)Basedonmodalparametersrecognitionofwholescreen,getlocationofdamagedsubstructure.(2)Locatefaultofsubstructureaccuratelybyacousticemissiontechnique.(3)Carryonresearchaboutfindingaefficientwaywecaninspectscreenforstructuredamageon-line.Theseprojectsaredonestepbystep.Atfirst,free-free
34、beamvibrationcharacteristicsarestudieddeeply.Thefirstrankandsecondrankbendingvibrationmodalshapeofbeamareabstractedasresearchobjects.ThestudyresultisshownthatthefrequenciesofFRFpeakvaluedrifttowardlowerfrequencyandtheamplitudesofFRFpeakvalueincreasewiththedamagedegreeofbeam.Thenthefirstrankandsecond
35、rankbendingvibrationcharacteristicsofbeamfixedonscreenarestudied.ThechangeregulationsofcharacteristicsofFRFwithbeamdamageareagreementtothatoffree-freebeam.ThereforethedamageinformationofbeamcanbegottenfromFREThewaveletpacketanalysismethodandspectralanalysiscalculationmethodareemployedinfrequencyresp
36、onseandtransmissibilityprocessing.Thefaultcharacteristicsareabstracted.Afterthen,thedamagedbeamhasaneffectonwholescreenvibrationcharacteristicsareresearched.Fromabovework,thedamagedbeamofscreencouldbelocatedfromwholescreen.Thentheacousticemissiontechniqueisusedtolocatefaultofthedamagedbeamaccurately
37、.Becausethetoomanysensorscouldntplaceontheworkingscreensothatwemustfindlimitedpositionstoplacesensorsgettingenoughstructuredamageinformation.Atlast,themethodofthefindingoptimumplacestoplacedsensorsforlocationwhichbeamisdamagedisstudied.Theefficientwayofoptimumplacetoplacedsensorsisfound.Inthisthesis
38、,thedifferentspectralanalysiscalculationmethodsareemployedinvibrationsignalprocessingtoabstractfaultcharacteristics.Theprocessingresultindicatesthatmethodsofvibrationsignalprocessionareefficiently.Thisthesisprovidessomerealizablewaystorealizetheon-linedynamicallyinspectscreenforstructuredamage.Trans
39、lation:射线探测焊缝及机械损伤方法概述图像中运动小目标的检测一直是图像处理与分析领域中的难题，它涉及到很多领域，具有很广泛的研究价值和应用价值。在工业探伤领域，由于焊接过程出现的各种问题，会导致焊缝中含有气孔和未焊透等缺陷，影响产品的质量和安全,所以焊接图像中缺陷目标的检测十分重要。目前X射线无损探伤系统主要采用人工方式进行在线检测与分析，而人工检测存在主观标准不一致、劳动强度大等缺点。因此，急需要研究一种有效的缺陷自动检测方法来代替人工检测，从而使在线检测工作客观化、规范化和智能化。本文的研究工作就是基于此而展开的。本文探讨了焊缝图像中缺陷目标的自动检测方法，主要做了以下几个方面的研究
40、:（1）针对X射线焊缝检测图像中存在大量与缺陷检测无关的背景冗余信息，采用了一种基于自适应闭值分割的焊缝区域的自动提取方法，以减少计算量，提高检测精度，取得了较好的效果。（2）由于SUSAN算法具有良好的抗噪能力，对图像的边缘、角点能够很好的识别，所以本文研究了一种以SUSAN算法为基础的，焊缝缺陷自动检测算法，同时辅助以形态学去噪和填充等运算，取得了较好的效果。（3）因为小波分析方法具有很好的局部化特性，它能对高频采取逐渐精细的时域或空域步长，从而可以聚焦到分析对象的任意细节。所以研究了一种利用小波分解来得到缺陷目标的形状和位置信息，并结合维纳滤波和形态学运算的焊缝缺陷检测方法，结果比较理想
41、。为了验证本文提出的两种算法的有效性，本文对在工厂实际得到的含有缺陷目标的焊接图像进行了检测，取得了较好的效果，证明了本文方法的可行性。焊管缺陷的自动检测是保证钢管产品质量的重要环节。在钢管生产过程中需要对焊管焊缝进行无损检测，X射线实时成象检测是一种比较有效的检测手段。随着生产率的不断提高，对焊管焊缝的自动化X射线检测要求越来越迫切，实现焊管焊缝的自动化检测具有重要的理论意义和实际意义。小波变换作为信号和图象处理的一种强有力的工具，其理论框架己基本形成，但从工程应用的角度，小波变换技术还处于初级阶段，还需进一步完善。.本文主要研究小波分析技术如何用于焊缝图象处理与识别以及一些相关技术。为建立
42、焊管焊缝自动定位检测控制系统，提出了焊缝图象识别的多机系统结构方案，该系统采用多处理器并行处理的结构，并由上位机工PC协调控制管理。系统能完成对焊缝图象的存贮、焊缝定位、缺陷识别和质量评定。并成功地研制了基于工SA总线的WatCh-Dog接口板，使用冗余法进行断点数据存储和恢复，实现了系统的可靠运行。硬件系统使用了Taxax仪器公司的高速信号处理器TMS320C30。图象采集卡能每秒连续采集25帧焊缝图象，使得实时完成焊缝图象处理与识别成为可能。在充分研究多分辨分析理论和分析信号处理器技术特点的基础上，针对DSPTMS320C3X的特点，提出了一种有限序列的FWT（快速小波变换）的改进算法，详
43、细阐述了信号处理器上FWT的周期性扩展的实现问题，用DSPTMS320C3X汇编语言实现了改讲的FWT算法。通过对小波变换系数进行阂值处理，给出了一种基于二维离散小波变换的图像去噪方法并用算子的形式加以描述。讨论了几种阂值选取方法和阂值策略，并提出了一种鲁棒局部闭值去噪法。用均方差衡量去噪性能，实验表明用鲁棒局部闭值去噪法好于全局闭值去噪法。为提高边缘检测的准确性和实时性，需要寻找一种适合于低对比度模糊焊缝图象边缘检测的快速方法。本文分析了焊缝图象的主要噪声来源及噪声与信号在小波变换下呈现的不同特点，提出了一种基于小波变换的多分辨率边缘检测方法。实验表明该算法的边缘检测效果明显优于经典的边缘检
44、测方法。针对具体的钢管直管对接焊缝图象，研究了其椭圆成象的几何关系，提出了焊缝图象区域模型，为进一步处理焊缝图象提供了理论依据。提出了一种模型基多分辨率图象自适应分割算法。该算法以OtSU判别准则为基础，结合焊缝区域模型进行焊缝图象的自适应目标分割，具有较好的分割效果。研究了在焊缝图象中目标缺陷的自动识别这一难题，在借鉴人类视觉系统知觉组织过程的基础上，提出了一种基于知识的多特征融合多窗口结构多分辨率目标识别算法。该算法依据一定的先验知识和准则，辅以人工智能的手段，能够得到较为精确的目标识别结果，是一种极有前途的智能识别算法。提出了一种几何特征快速提取算法，该算法改变了通常先图象二值化后提取目
45、标参数特征的做法，而是直接对灰度图象进行目标参数特征提取。使用本文提出的几何特征快速提取算法可以有效地实现缺陷目标的快速识别处理。随着生产的发展与科学技术的进步，各类机械设备和结构向着大型、高效化发展，从而也使得这些机械设备和结构趋于复杂化。复杂的承载条件与长时间的连续工作，导致设备结构的损伤不可避免。因断裂、疲劳、腐蚀和磨损而造成的破坏，其损失达美、日等国家每年国民经济总值的6%,8%o而在我们国家20世纪80年代的结构损伤事故率比工业化国家高10倍，人员累计伤亡居国内劳动安全事故第二位。1986年，美国的“挑战者，号航天飞机失事损失高达12亿美元;苏联的切尔诺贝利核电站的核泄漏事故，对整个
46、地区的人员、生态环境都是无法估量的损害。1985年我国大同电厂一机组联轴器断裂事故、1988年秦岭电厂机组主轴断裂，造成的经济损失均近亿元，并严重影响华北和西北地区供电。从1984年到1991年，我国50MW以上的汽轮发电机组转子严重损坏等重大事故就达6起。因此，研究防止这类事故发生的根本途径一一大型复杂机械结构的健康状况监测与故障诊断（尤其是早期故障诊断）的理论和技术，实现结构损伤的早期识别，及时采取措施，防止损伤的发展，以保证这些系统安全、可靠、长寿命、高效率地运行成为紧迫的任务。然而，对于大型的复杂机械结构，实现合理、优化地布置传感器，监测其运行结构的局部损伤，从中识别出损伤的结构件及损
47、伤状态，进行早期故障预报是具有相当的难度的。国内外的研究报道显示，由于结构振动模态参数对不同损伤各有其敏感性，加之大型复杂结构振动模态识别技术发展的限制，以振动特性为参数对结构进行损伤检测的研究大多集中在诸如直梁、析架等简单对称结构，且多是基于有限元分析数值计算得出相应的结论。而实际设备的构造及联接都是相当复杂的，无法实现可靠的有限元建模。如何实现用振动测试技术对实际复杂上业结构的健康状况进行监测与故障的在线诊断还是一个急待解决的问题。故障诊断技术具有很强的学科交叉性，尤其是现代控制理论，信号处理，模式识别、最优化方法、决策论、人工智能等的迅速发展，为解决复杂系统的故障诊断问题提供了理论基础，
48、形成了许多具体的方法。故障信号的特征提取为故障准确诊断的前提条件。近三十年来，各类机械设备基于实时监测（包括振动监测）的故障诊断技术的研究和应用，促进了故障信号处理和特征提取技术的发展。这些技术包括时域信号波形分析和统计特征值提取;基于FFT分析的高斯平稳随机信号现代谱分析技术咱功率谱和互功率谱、高阶谱、倒谱、复倒谱以及谱嫡与极大嫡谱估计（也包括与之相对应的自相关与互相关函数、高阶自相关函数等时域分析）；非平稳信号的时频分析和多元统计分析等。故障诊断技术还涉及到材料的选择、制造工艺、结构设计、断裂力学等多种学科和专业技术领域。随着力学、材料科学、物理学、化学领域的学科交叉与发展，可从缺陷的背景
49、和损伤、断裂机制来研究从材料变形、损伤到失效的全过程。而计算机数据处理、模式识别的技术发展为与早期故障相关的微弱信号的捕捉和提取，提供了有利的手段。结构损伤检测是一个具有广阔工程应用背景的研究课题，而大型复杂结构损伤检测的研究离实际应用还有距离，还有一许多问题需要在今后的研究中加以解决:如精确的测量信息处理技术的发展，以期获得更加精确的测量模态;基于吸收各种方法优点的混合模型的结构损伤检测方法的研究;大型设备的工况监测，传感器的合理布置与优化配置问题的具体应用;其它理论方法的引入，如模糊数学的应用以及子结构振动分析方法的应用等等。本课题以大型振动机械为研究对象，进行结构损伤检测方法研究。常规的无损检测方法，如超声波、射线以及磁力探伤与渗透法探伤等大多是用于生产过程中间环节的零件离线检测和设备检修，通常为静态检测。本项目研究的基本思路是对大型振动机械进行多测点建模，利用振动测试技术进行充分的离线试验和分析来获取被诊断结构的振动特性细节、故障机理及其特征，作为结构动力学本质特征库的先验知识与在线自动故障诊断信息库的重要内容;研究以结构的少测点获取结构损伤信息的建模方法，合理配置传感器，
